THPrompt

We provide the code (in PyTorch) for our paper "Efficient Prompt Tuning for Spatial-Temporal Graphs under Topology and History Shift".

Usage

1. Download Data

From JODIE dataset website, download the required data files (e.g., wikipedia.csv, reddit.csv). Place the downloaded files into the processed/ directory.

2. Process Data

Run the following scripts to preprocess the raw data.

# The --data argument can be changed to 'reddit' or other datasets
python utils/preprocess_data.py --data wikipedia --bipartite
python utils/downstream_process.py

The pre-training part of the code references the paper "Node-Time Conditional Prompt Learning in Dynamic Graphs".

3. Pre-train the Model

Execute the following command to pre-train the model.

python pre-training.py --use_memory

4. Run Downstream Task

Use the pre-trained model to perform the downstream task.

python downstream_task.py --use_memory